Why it exists
The project started in 2019 because a handful of university labs and small hosting providers kept hitting rate limits on the large public archives during release weeks. Rather than each of them running a half-maintained local copy, we put one properly monitored mirror in front and pointed everyone at it.
Nothing about that has changed. The archive is not a CDN, it does not have an SLA, and it is
not trying to grow. It exists so that a apt update in a classroom in the
Netherlands does not fail because somebody in another country decided to reinstall their
laptop at the same time.
How it is built
A single instance pulls from upstream with rsync, verifies signatures, and swaps
the published tree atomically. There is no origin pull, no caching layer and no rewriting
proxy in front of the files — what you download is what the upstream published.
- Compute
- 1 vCPU (Ampere), 6 GB RAM
- Storage
- Local NVMe, 46 GB working set with hourly rotation of release ISOs
- Location
- Amsterdam, Netherlands
- Web server
- Caddy, TLS terminated locally, HTTP/2 and HTTP/3
- Sync
- rsync 3.2, bandersnatch 6.x, athens for the Go proxy
- Monitoring
- Blackbox probes every 60 s, checksum sampling weekly
Usage policy
Three rules, in order of how often they are broken:
- Do not mirror this mirror. Pull from upstream. Re-mirroring adds load and gives you stale metadata.
- Respect the concurrency limit. 40 connections per source address. Tools that open a connection per file will be shaped, not blocked.
- Do not attempt to make this a general-purpose file host. The archive serves the trees listed on the mirror page and nothing else.
There is no authentication and no account system. If your use case needs credentials, this is the wrong service.
What we log
Aggregate counters only: requests per hour, bytes served per tree, error rates. Per-request logs are kept for 48 hours for debugging, with the last octet of the client address removed before anything is written. No cookies, no analytics scripts, no third-party fonts or CDNs on this site.
Funding
Hosting is covered by credits from a cloud provider programme and by three individuals who pay for the storage out of pocket. There is no advertising and no sponsorship. If the credits run out, the mirror will be announced as closing rather than monetised.
Who
Four maintainers, rotating on-call. We do not publish names, mostly because the work is unglamorous and partly because we would rather people send problems to the role address than to individuals. Reach the team via the contact page.